AI workflow

Building got cheap.
Knowing it landed
did not.

Spend is easy to count. Landing is not.

Agents now do the read, the draft, and the check, so a team ships more and the bill climbs with it. Neither of those numbers says whether the work changed anything. A team of people and agents needs one live surface across strategy, design, the build, and go-to-market, where a person can see effort on one side and outcome on the other and decide where the next lap of effort goes. The record that surface runs on is already there, scattered across every tool the team works in.

The gap

The bill compounds.
The work that lands doesn't follow it.

Agent spend (last 30 days)
142M+38%vs last month
Changes shipped with agent help
12±0vs last month
◆ The scissor · Agent spend vs changes shipped · 90 days
100Agent spend138 · +38%Changes shipped100 · ±0T-5moT-4T-3T-2T-1Today

The bill compounds. The shipped work holds flat. I call this the scissor, and I keep meeting it in cost, in the quality of what teams write down, and in build vs buy. The line neither of these draws is the third one: did any of it land?

Illustrative, drawn from shapes I keep running into. Not a customer read. The specifics differ. The pattern rarely does.

Where the read comes from

Two streams,
from tools the team already works in.

The frameworks do the reading. Writing code and making design both leave a trail: how the work went, what it cost, how many passes it took. Put in one shared vocabulary, the two streams show where the effort actually went. Sample lists below, not a customer's numbers.

In practice that is the coding tool and the design tool, read the same way: how many passes a change took, what it cost to get there, and whether what it produced survived review.

What it settles

Five questions,
answered by the same trail.

These land on the same scale as everything else in the system, so they read against the same ladder and need no vocabulary of their own. Each one is a question a person has to settle, with the evidence sitting next to it.

  1. What gets written down

    Whether the team explains the same context to an agent every session, or writes it down once and moves on.

  2. What it costs

    Where the spend really goes: which person, which codebase, which task. Not just the monthly bill.

  3. Build or buy

    Where agent effort turns into work that landed, and where it only looks busy.

  4. How the work runs

    Which habits before building actually shorten the trip to something shipped.

  5. What the team knows

    Who reaches for which saved workflows and connected tools, how often, and to what effect.

Illustrative read

Pattern:
context debt.

Spend is climbing, planning before building is falling, and the team is drifting apart on which saved workflows it uses. Named and put in front of one person, the pattern points at the few moves that close it.

◆ PatternIllustrative
Claude CodeClaude Code
7-day window12 contributors
Tracking

Spend is up 38%. Planning before building is down 6 points.

Context debtHabits slippingUneven tool use
The read · illustrative

The read: This team is spending more per task without the written-down context that would make the spend pay. The bill scales. The work that lands does not. Three moves close it, and one person owns each.

  1. Plan before building, by default, across the teamHow the work runsMeasured
  2. Write down the three workflows the team repeats mostWhat the team knowsProjected
  3. Send routine edits to the smaller, cheaper modelWhat it costsProjected

Naming the pattern is the easy part. Catching it while there is still a quarter left to act on it takes one live surface and one person willing to make the call.